A Personalized Movie Recommendation Algorithm Incorporating The User's Interests And Emotional Features

With the rapid increase in the amount of network information,traditional information retrieval methods have begun to encounter some performance bottlenecks.For example,the retrieval for content information requires users to provide clear demand,and the search method using keywords cannot filter massive retrieval results.The appearance of intelligent recommendation systems can alleviate the problem of information overload to a certain extent and improve the efficiency of information use.In addition,recommendation systems are key tools to improve user experience and promote sales/services for many online websites and mobile applications.For example,80% of movies watched on Netflix came from recommendations,and 60% of video clicks on YouTube came from users' homepage recommendations.The personalized movie recommendation systems recommend movies according to users' diverse preferences,which can not only save a lot of their time for information search,but also establish a close relationship between websites and users,guiding them to rely on the recommendation so as to form a virtuous cycle.For movie recommendation websites or video websites,accurate recommendation results can increase users' click rate and purchase rate,thus bringing considerable profits to enterprises.This paper mainly researches on the following three aspects:First,building a hybrid movie recommendation model based on ratings and reviews.The proposed recommendation model can be divided into(1)A personalized recommendation algorithm incorporating the user's interests and emotional features(SentiNMF-U);and(2)A personalized recommendation algorithm incorporating the movie's topic characteristics and the user's emotional features(SentiNMF-I).This paper extracts the user's interest topic and movie's feature topic,which not only highlights the characteristics of each user and each movie,but also effectively alleviates the problem of data sparsity in the recommendation system.Second,finely dividing the film reviews according to the sentiment orientation of the sentences.Before utilizing the topic model to extract the topic of the movie review,this paper divides the review into positive ones and negative ones according to the sentence sentiment analysis.This process can more vividly depict the user's and movie's profile,making the recommendation strategy avoid recommending similar negative preferences and more accurately approaching the user's positive preferences.Third,adopting a “voting mechanism” based on collaborative filtering to generate a recommendation list.This paper creatively uses a “voting mechanism” that assigns voting rights according to similarity to form a recommendation set,which can avoid the problem of weight solving in the traditional collaborative filtering recommendation.The experimental results on the real Douban movie data set show that when the data set is very sparse(i.e.most users on the movie website only give five or less ratings and reviews),the two proposed algorithms in this paper can achieve better results than the most representative recommendation algorithm.Moreover,when comparing the proposed algorithms,this paper finds that the user-based algorithm is superior to the movie-based algorithm,and it also proves that it is meaningful to divide movie reviews into positive and negative ones according to user's sentiment orientation when calculating the topic distribution.In view of data sparsity in personalized recommendation systems,this paper provides new solutions and methods verified on a real data set,which has good enlightenments for relevant scholars and researchers in this field.